2020 Volume 49 Issue 2 Pages 136-143
Recently, cancer is the leading cause of death in Japan. It is known that early detection and early treatment of cancer before metastasis occurs is important. Therefore, circulating tumor cells (CTC) is expected as a useful biomarker that can be new cancer tests. The CTC exists in the blood of patients with metastatic cancer and pathologists analyze CTC to diagnosis the condition of cancer. Pathologists analyze blood samples from images taken with a fluorescence microscope, but it is time-consuming since the number of CTC in the blood is very few. In this paper, we develop an automatic detection method of CTC in fluorescence microscopy images. In the proposed method, we detect cell regions by using the selective enhancement filter and blob analysis and then identify CTC by using SqueezeNet, which is one of the kinds of convolutional neural network (CNN). The input image to SqueezeNet is a composite of three images taken with a fluorescence microscope. As a result of applying the proposed method to 5,040 microscope images (6 cases), 97.30[%] of true positive rate (TPR) and 3.150[%] of false positive rate (FPR) was obtained.
